A Costophrenic Angle Estimator in Frontal Chest Radiographs

Abstract

Automatic detection of Tuberculosis (TB) is gaining popularity nowadays due to the lack of trained medical personnel (radiologists) and the limited cost and high performances achieved by current software applications. The task is difficult and challenging due to the various manifestations of the TB such as cavitations, infiltrates, opacities, hilar enlargements, consolidations, etc. These textural anomalies are difficult to detect and the training of such a system needs a lot of training material. However, the blunted costophrenic angle is quite a visible cue and its detection can lead directly to the right diagnoses without any expensive image processing and machine learning operation. The experiments conducted on the Montgomery dataset [1] show a variance less than 3.9 degrees which are way better than previously proposed solutions, - proving the efficiency and the robustness of the proposed method.